Numerical Techniques for Shallow Water Waves
-
Upload
nilim-sarma -
Category
Documents
-
view
225 -
download
2
Transcript of Numerical Techniques for Shallow Water Waves
Version 1.2
Numerical Techniques for the Shallow Water Equations
������ ����
The University of Reading, Department of Mathematics, P.O.Box 220,
Whiteknights, Reading, Berkshire, RG6 6AX, UK
E-mail: [email protected]
Numerical Analysis Report 2/99
Abstract
In this report we will discuss some numerical techniques for approximating the
Shallow Water equations. In particular we will discuss finite difference schemes,
adaptations of Roe’s approximate Riemann solver and the Q-Schemes of
Bermudez & Vazquez with the objective of accurately approximating the solution of
the Shallow Water equations. We consider four different test problems for the
Shallow Water equations with each test problem making the source term more
significant, i.e. the variation of the Riverbed becomes more pronounced, so that
the different approaches discussed in this report can be rigorously tested. A
comparison of the different approaches discussed in this report will also be made
so that we may determine which approach produced the most accurate numerical
results overall.
The work contained in this report has been carried out as part of the Oxford / Reading
Institute for Computational Fluid Dynamics and was funded by the Engineering and Physical
Science Research Council and HR Wallingford under a CASE award.
1
1 Introduction
Throughout this report, we will be discussing some numerical techniques for
approximating the Saint-Venant equations, i.e.
( ) ( )wRwFw
x,xt
=∂
∂+∂∂
(1.1)
where
( ) ⎥⎦
⎤⎢⎣
⎡=
uh
htx,w , ( )
⎥⎥⎦
⎤
⎢⎢⎣
⎡+= ghhu
uh22
2
1wF and ( ) ( )⎥⎦⎤
⎢⎣
⎡′
=xHgh
x0
, wR .
Here, h(x,t) and u(x,t) represent the water depth and the fluid velocity respectively and
H(x) is the bed depth from a fixed reference level, see Figure 1-1.
Figure 1-1: Shallow Water Variables.
D
L
x
H(x)
h(x,t)
Riverbed
River
B(x)
2
Notice that (1.1) has a source term present, ( )wR ,x , which can cause difficulties in
accurately approximating (1.1), especially if the source term is stiff (see Hudson[7]
and LeVeque & Yee[9]). Sometimes, it is convenient to re-write the source term in
terms of the bed height
( ) ( )xBDxH −= ⇒ ( ) ( )xBxH ′−=′ ,
hence,
( ) ( ) ( )⎥⎦⎤
⎢⎣
⎡′−
=⎥⎦
⎤⎢⎣
⎡′
=xBghxHgh
x00
,wR .
In Chapter 2, we will discuss four test problems for the Shallow Water Equations
(1.1) where the source term becomes more significant for each test problem, i.e. the
variation of the riverbed becomes more pronounced. The first test problem will have
no source term present and the second and third test problem will have a source term
present, with the third test problem having a source term more significant than the
second test problem. The final test problem will also have a source term which is the
most significant out of all of the four test problems and thus is the most difficult to
approximate accurately.
The four test problems will allow us to rigorously test the different numerical
approaches for approximating the Shallow Water Equations (1.1), which are discussed
in Chapter 3, and determine if the approaches are accurate when the source term
becomes difficult to approximate accurately. The numerical approaches we will
discuss in Chapter 3 are the finite difference approach, adaptations of Roe’s
approximate Riemann solver and the Q-Schemes of Bermudez & Vazquez[1]. Some
of these approaches require the Jacobian matrix of ( )wF
⎥⎦
⎤⎢⎣
⎡−
=∂∂=
uugh 2
10)(
2wF
wA ,
3
which has eigenvalues
ghu +=λ1 and ghu −=λ2
and eigenvectors
⎥⎦
⎤⎢⎣
⎡+
=ghu
11e and ⎥
⎦
⎤⎢⎣
⎡−
=ghu
12e .
All of the numerical approaches discussed will derive numerical schemes that are
either first order, second order or flux-limited second order schemes (see
LeVeque[10], Kroner[8] and Sweby[14]). In Chapter 4 the different numerical
approaches discussed in Chapter 3 will be compared by using the four test problems
so that we may determine which approach produced the most accurate numerical
results overall.
4
2 Test Problems
In this chapter, we will discuss four test problems, for the Shallow Water Equations
(1.1). The first test problem is the dam-break problem and was discussed by
Glaister[3] and Stoker[13]. This problem is one of the most basic problems for (1.1)
since no source term is present. The second test problem contains a source term and
represents a dam breaking on a variable depth riverbed. The third problem was
discussed by LeVeque[11] and represents a square pulse, which breaks up into two
waves travelling in opposite directions, on a variable depth riverbed. The final test
problem was discussed by Bermudez & Vazquez[1] and represents a tidal wave
propagating on a variable depth riverbed. For each test problem, notice how the
source term becomes more significant, i.e. the variation of the bed depth becomes
more pronounced.
2.1 Problem A - The Dam-Break Problem
In this test problem, (1.1) has no source term present, i.e.
( )0=
∂∂+
∂∂
xt
wFw. (2.1)
This is due to the riverbed being of constant depth resulting in ( ) 0=′ xH ∀x. We also
have initial conditions
u(x,0) = 0 and ( )⎪⎪⎩
⎪⎪⎨
⎧
≤<φ
≤≤=
12
1 if
2
10 if 1
0,
0 x
xxh ,
5
which are illustrated in Figure 2-1. In Figure 2-1, the discontinuity at x = 0.5
represents a barrier, which separates the two initial river heights and is removed at t =
0. Walls are present at x = 0 and at x = 1 resulting in reflection at both boundaries.
Note that, for this test problem, if 2.71
0
>φ
then both eigenvalues of )(wA are of the
same sign and the downstream flow is supercritical. If 2.71
0
<φ
then the eigenvalues
of )(wA are of opposite sign and the downstream flow is subcritical. If the
downstream flow is supercritical then difficulties can arise in accurately numerically
approximating the wave speed of the discontinuity at x = 0. In Figure 2-1,
2.725.0
11
0
<==φ
, hence the downstream flow is subcritical.
We may obtain an exact solution of the dam-break problem by using the analysis of
Stoker[13]. The following exact solution is illustrated in Figure 2-2 and Figure 2-3
and is only valid for φ0 = 0.5 and φ1 = 1,
( )( ) ( )
( )
⎪⎪⎪⎪
⎩
⎪⎪⎪⎪
⎨
⎧
+>
+≤<+−
+−≤≤−+−
−<
=
2
1S if 0
2
1S
2
1 if
2
1
2
1 if 212
3
12
1 if 0
,
222
22
tx
txtcuu
tcuxgt gtxt
gtx
txu
and
( )( )
( )
⎪⎪⎪⎪⎪
⎩
⎪⎪⎪⎪⎪
⎨
⎧
+>
+≤<+−⎟⎟
⎠
⎞⎜⎜
⎝
⎛−+
+−≤≤−⎟⎠⎞⎜
⎝⎛ −−
−<
=
21
S if 21
21
S21
if 116
141
21
21
if 2
122
91
21
if 1
,
22
2
22
2
tx
tx tcug
S
tcuxgt t
xg
g
gtx
txh
6
where
⎟⎟⎠
⎞⎜⎜⎝
⎛++−=
gS
S
gSu
16118
2
2 ,
⎟⎟⎠
⎞⎜⎜⎝
⎛−+= 1161
4
2
2gSg
c
and the wave speed of the discontinuity created at x = 0 is
S = 2.957918120187525.
For a more in depth analysis on how the value of S was obtained see Glaister[3] and
Stoker[13].
2.2 Problem B - The Dam-Break Problem on a Variable
Depth Riverbed
This test problem is similar to Problem A but the riverbed is no longer of constant
depth, resulting in ( ) 0≠′ xH for some values of x, which means that a source term is
present, i.e.
( ) ( )wRwFw
,xxt
=∂
∂+∂∂
.
For this test problem, the riverbed is defined as
⎪⎩
⎪⎨
⎧≤≤⎟⎟⎠
⎞⎜⎜⎝
⎛+⎟
⎠⎞
⎜⎝⎛
⎟⎠⎞
⎜⎝⎛ −π
=
otherwise 0
5
3
5
2 if 1
2
110cos
8
1)(
xxxB
and we have initial conditions
u(x,0) = 0 and ( )( )
( )⎪⎪⎩
⎪⎪⎨
⎧
≤<−φ
≤≤−=
12
1 if
2
10 if 1
0,
0 x xB
xxBxh ,
7
which is illustrated in Figure 2-4. In Figure 2-4, the discontinuity at x = 0.5 represents
a barrier, which separates the two initial river heights and is removed at t = 0. Walls
are again present at x = 0 and at x = 1 giving reflection at these boundaries. Since a
source term is now present, difficulties can arise when numerically approximating this
problem, as we will see later.
2.3 Problem C - A Problem Discussed by LeVeque[11] In this test problem, the riverbed is again of variable depth, resulting in ( ) 0≠′ xH for
some values of x, which means that a source term is present, i.e.
( ) ( )wRwFw
,xxt
=∂
∂+∂∂
.
For this test problem, the riverbed is defined as
⎪⎩
⎪⎨
⎧≤≤⎟⎟⎠
⎞⎜⎜⎝
⎛+⎟
⎠⎞⎜
⎝⎛
⎟⎠⎞
⎜⎝⎛ −π
=
otherwise 0
5
3
5
2 if 1
2
110cos
4
1)(
xxxB
and we have initial conditions
u(x,0) = 0 and ( )( )( )
( )⎪⎩
⎪⎨⎧
>−≤≤−
<−=
0.2 if 1
0.20.1 if 1.2
1.0 if 1
0,
xxB
xxB
xxB
xh ,
which is illustrated in Figure 2-5. This problem represents an initial pulse, which
breaks up into two waves moving in opposite directions. The right-going square-
wave pulse passes the hump in the riverbed and becomes partially reflected, causing a
disturbance behind the hump. In this problem, there are no walls present at x = 0 and
x = 1 and reflection does not occur.
8
2.4 Problem D - Tidal Wave Propagation on a Variable
Depth Bed
This test problem was discussed by Bermudez and Vazquez[1] and also has a source
term present, i.e.
( ) ( )wRwFw
,xxt
=∂
∂+∂∂
.
We have a variable depth given by
⎟⎠⎞⎜
⎝⎛
⎟⎠⎞⎜
⎝⎛ +π+−=
2
14sin10
405.50)(
L
x
L
xxH ,
with initial conditions
u(x,0) = 0 and h(x,0) = H(x),
which are illustrated in Figure 2-6, and boundary conditions
⎪⎩
⎪⎨
⎧
>
≤⎟⎟⎠⎞
⎜⎜⎝⎛
⎟⎠⎞⎜
⎝⎛ −+
=43200 if 5.60
43200 if 21
864004
sin45.64),0(
t
tt
thπ
,
representing an incoming wave which is illustrated in Figure 2-7, and u(L,t) = 0. For
this test problem, the reference depth D is taken as D = 60.5.
This problem represents a tidal wave propagating on a variable depth riverbed. Here,
h(0,t) represents a tidal wave of 4m amplitude and in Figure 2-7, we can see that at t =
21,600s, the tidal wave has reached it’s full height of 8m at inflow whereas at t =
43,200s, the tidal wave has disappeared. Also, since the waves propagate at speed
gh , the tidal wave should only reach as far as 216,000m at t = 10,800s, when L =
648,000m.
Throughout this chapter, we have discussed four different test problems for the
Shallow Water Equations. In the next chapter, we will discuss some numerical
9
techniques for approximating the Shallow Water Equations so that we can apply the
different numerical techniques to the four test problems.
Initial Conditions for Problem A
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
h(x,
0)
Figure 2-1: Initial condition h(x,0) for Problem A
10
Exact Solution of Problem A
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
h(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 2-2: Illustration of the exact solution of Problem A
E xact S olution of P roblem A.
0
0 .1
0 .2
0 .3
0 .4
0 .5
0 .6
0 .7
0 .8
0 .9
1
0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8 0 .9 1
x
u(x,
t)
0 0 .01 0 .02 0 .03 0 .04 0 .05 0 .06 0 .07 0 .08 0 .09 0 .1 Figure 2-3: Illustration of the exact solution of Problem A
11
Initial Conditions for Problem B
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
B(x) h(x,0)+B(x)
Figure 2-4: Initial condition h(x,0) and B(x) for Problem B.
Initial Conditions for Problem C
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
B(x) h(x,0)+B(x)
Figure 2-5: Initial condition h(x,0) and B(x) for Problem C.
12
Initial Conditions for Problem D
0
10
20
30
40
50
60
70
0 2000 4000 6000 8000 10000 12000 14000
x
RiverBed River
Figure 2-6: Initial condition h(x,0) and B(x) for Problem D.
Boundary Condition h(0,t) for Problem D.
0
1
2
3
4
5
6
7
8
9
10
0 10800 21600 32400 43200 54000 64800
t
60.5
- h
(0,t)
Figure 2-7: Boundary condition h(0,t) for Problem D.
13
3 Numerical Schemes
There are a variety of numerical techniques for approximating (1.1), e.g. finite
element methods, finite volume methods, etc. In this report, we will discuss the finite
difference approach (see LeVeque[10] and Kroner[8]), adaptations of Roe’s
approximate Riemann solver (see Glaister[4], Hubbard[6] and Roe[12]) and the Q-
Schemes of Bermudez & Vazquez[1].
These approaches will be used to derive first order and second order numerical
schemes where if a numerical scheme is first order then the scheme is dissipative and
if a numerical scheme is second order then the scheme is dispersive (see Figure 3-1).
Dissipation occurs when the travelling wave’s amplitude decreases resulting in the
numerical solution being smeared. Dispersion occurs when waves travel at different
wave speeds and results in oscillations being present in the numerical results. Both
dissipation and dispersion can cause very significant errors in the numerical results,
see Figure 3-1, and can sometimes give completely inaccurate numerical results.
One way to minimise dissipation and dispersion is to use a numerical method which
satisfies the Total Variational Diminishing property (see Sweby[14] and Harten[5]).
Flux-limiter methods satisfy the TVD property and switch between a second order
approximation when the region is smooth and a first order approximation when near a
discontinuity. Flux-limiter methods will also be applied to the different numerical
approaches so that oscillations present in the numerical solution can be minimised.
14
Numerical Results of an Advection Test Problem using First and Second Order Finite Difference Schemes with the Exact Solution.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
0.5)
Exact Solution First Order Scheme with Dissipation present Second Order Scheme with Dispersion Present
Figure 3-1: Illustration of Dispersion and Dissipation
Figure 3-2: The Mesh
x = x0
i = 0 x = x1 i = I
x x1 x2 x3 …. xI-3 xI-2 xI-1 xi-1 xi xi+1 ……
n = 0 t0 = 0
u(x,0)
tn-1
t1
t2
…
Δt tn
tn+1
uni un
i 1+ uni 1−
uni
11++ un
i1+ un
i11+−
uni
11−− un
i1− un
i11−+
..….
Δx
t
n = N t = tN
tN-1
tn+1 – tn = Δt xi+1 – xi = Δx
tn = nΔt xi = iΔx
tN-2
15
3.1 Mesh & Boundary Conditions Before we discuss the various numerical approaches for approximating the Shallow
Water Equations (1.1), we must define the mesh and then look at the numerical
boundary conditions required to implement the numerical approaches correctly.
In this report, we will use a fixed mesh over the finite region xxx Io ≤≤ and tt N≤≤0 ,
which is illustrated in Figure 3-2. Here, the numerical solution is denoted by
( )tnxiuuni ΔΔ≈ , where xxx ii 1−−=Δ and ttt nn 1−−=Δ for all i and n.
Numerical boundary conditions are required at x = x0 and x = xI and are derived
separately for each test problem. For
i) Problem A and Problem B, walls are present at the upstream boundary, x
= 0, and the downstream boundary, x = 1, so we will need to reflect the velocity at the
two boundaries
uu ni
ni −=− and uu n
iIn
iI −+ −=
where i = 1,2. For the water depth, we will assume that the depth is constant at the
boundaries
hh nni 0=− and hh n
In
iI =+ where i = 1, 2.
ii) Problem C we will assume that the velocity and water depth are constant
hh nni 0=− , hh n
In
iI =+ , uu nni 0=− and uu n
In
iI =+
where i = 1, 2.
iii) Problem D we use the analytical boundary conditions to obtain
( )thhni ,0=− and 0=+un
iI
where i = 0, 1, 2. For the second boundary condition, we will use
hh iIn
iI0++ = and uu nn
i 0=−
where i = 1, 2.
16
3.2 Finite Difference Method One approach widely used to numerically approximate (1.1) is the finite difference
method. This method involves replacing the derivatives of (1.1) with finite difference
approximations, e.g.
tt
ni
ni
Δ−=
∂∂ + www 1
which is a forward difference approximation in time, to obtain a finite difference
scheme. Great care must be taken when using finite differences to construct a finite
difference scheme as we need to ensure that the scheme is conservative. A finite
difference scheme that is not conservative may propagate discontinuities at the wrong
wave speed, if at all, giving inaccurate numerical results. To ensure we obtain a
conservative scheme, we only construct finite difference schemes of the form
[ ]FFww *2/1
*2/1
1−+
+ −ΔΔ−= ii
ni
ni
x
t, (3.1)
where F* is called the numerical flux function. Notice that (3.1) approximates the
homogeneous problem (2.1). Obtaining a conservative scheme when a source term is
present can be very difficult, especially when the source term is stiff. However, there
are a variety of numerical techniques which approximate the source term and they
generally fall under two categories: a pointwise approach, where we add a source term
approximation to (3.1), i.e.
[ ] RFFww tx
t niii
ni
ni Δ+−
ΔΔ−= −+
+ *2/1
*2/1
1 (3.2)
or a physical approach, which involves taking an average value of the source term and
upwinding (see Sweby[15] for more details), i.e.
[ ] RFFww tx
tiii
ni
ni Δ+−
ΔΔ−= −+
+ **2/1
*2/1
1 . (3.3)
17
We will now look at specific schemes which numerically approximate the Shallow
Water Equations7.
3.2.1 First Order Lax- Friedrichs Approach We can use the Lax-Friedrichs scheme with a source term approximation ‘added’ to
approximate (1.1), i.e.
( ) RFFww
w ss n
ini
ni
ni
nin
i +−−+= −+−++
11111
22 (S-1)
where
x
ts
ΔΔ= , ⎥
⎦
⎤⎢⎣
⎡=
uh
hni
ni
nin
iw , ( ) ( ) ⎥⎥⎦
⎤
⎢⎢⎣
⎡
+=hguh
uhni
ni
ni
ni
ni
ni 2 2
2
1F and ( )⎥⎦⎤
⎢⎣
⎡−−
=−BBhg ii
ni
ni
1
0R .
Notice that this scheme is in a similar form of (3.2) and that the Jacobian matrix of
F(w) associated with the system (1.1) does not need to be approximated. However,
the Lax-Friedrichs scheme can suffer from oscillations but sometimes these
oscillations can be minimised by using sufficiently small step-sizes.
3.2.2 Explicit MacCormack Approach We can also use an approach discussed by LeVeque & Yee[9] to approximate (1.1),
which involves modifying the explicit MacCormack scheme, i.e.
( ) ( ) RFFwww )1()1(1
)1()1(1
222
1iiii
ni
ni
ss +−−+= −+ (S-2a)
where ( ) RFFww ss n
ini
ni
nii +−−= +1
)1( ,
x
ts
ΔΔ= , ⎥
⎦
⎤⎢⎣
⎡=
uh
hni
ni
nin
iw , ( ) ( ) ⎥⎥⎦
⎤
⎢⎢⎣
⎡
+=hguh
uhni
ni
ni
ni
ni
ni 2 2
2
1F and ( )⎥⎦⎤
⎢⎣
⎡−−
=−BBhg ii
ni
ni
1
0R .
Unfortunately, this scheme suffers from dispersion and results in oscillations being
present in the numerical approximation. However, LeVeque & Yee[9] eliminated
18
these oscillations by adapting (S-2a) so that the scheme satisfies the TVD property
and obtained
( )ϕϕ 2/12/12/1 2/1)2(1
−−+ ++ −+= iii ii
ni XXww (S-2b)
where X represents a matrix containing the right eigenvectors e, w )2(
i is the numerical
approximation derived from scheme (S-2a), i.e.
( ) ( ) RFFwww )1()1(1
)1()1()2(
222
1iiii
nii
ss +−−+= − ,
and
⎥⎦
⎤⎢⎣
⎡φφ
=+
++ 2
2/1
12/1
2/1
i
iiϕ
where
( )( )( )Qvvki
ki
ki
ki
ki 2/12/12/1
22/12/1 2
1+++++ −α−=φ ,
( )ααα= ++−+ki
ki
ki
kiQ 2/32/12/12/1 ,,modmin ,
( ) ( ) ( ) ( ) ( )⎩⎨⎧ ===
=otherwise 0
sgnsgnsgn if ,,min,,modmin
cbadcbadcba ,
( )wwX ni
nii
i
i −=⎥⎦
⎤⎢⎣
⎡αα
+−+
+
+1
12/12
2/1
12/1 and λΔ
Δ= ++ki
ki
x
tv 2/12/1 .
Here, k represents the kth component of the vector and λ, e and α represents the
eigenvalues, eigenvectors and wave strengths associated with the system (1.1)
respectively, which can be obtained by the decomposition
wAeF Δ=∑ λα=Δ=
~~~~2
1k
kkk .
Variables which have ˜ represent the Roe average and A~
is the Jacobian matrix
evaluated at Roe’s average state. We will discuss the method of decomposition in
Section 3.3.
19
3.2.3 Semi-Implicit MacCormack Approach LeVeque & Yee[9] also obtained a semi-implicit approach to approximate (1.1),
which involves modifying the explicit MacCormack scheme so that
( )ϕϕ 2/12/12/1 2/1)2(1
−−+ ++ −+= iii ii
ni XXww (3.4)
where
( ) ( ) RwR
IFFwR
Iwww )1(
1
)1(1
)1(
1
)1()2(
22222
1iiii
nii
ssss⎥⎦⎤
⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−+−⎥⎦
⎤⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−−+=
−
−
−
,
( ) RwR
IFFwR
Iww ni
ni
ni
nii
ss
ss ⎥⎦
⎤⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−+−⎥⎦
⎤⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−−=
−
+
−
22
1
1
1
)1(
and
( ) ⎥⎦
⎤⎢⎣
⎡−−
=⎥⎦⎤
⎢⎣⎡∂∂
− 0
00
1BBg ii
n
iwR
which implies that
( ) ⎥⎥⎦
⎤
⎢⎢⎣
⎡−−=⎥⎦
⎤⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−
−
−
12
01
2 1
1
BBgss
iiwR
I .
Also, notice that for the Shallow Water Equations
( ) ( ) ( ) RRwR n
iii
niii
niii
ni
BBhgBBhgBBgss =⎥
⎦
⎤⎢⎣
⎡−−
=⎥⎦
⎤⎢⎣
⎡−−⎥
⎥⎦
⎤
⎢⎢⎣
⎡−−=⎥⎦
⎤⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−
−−−
−
111
100
12
01
2I .
(3.5)
Hence, we can re-write (3.4) as
( )ϕϕ 2/12/12/1 2/1)2(1
−−+ ++ −+= iii ii
ni XXww (S-2c)
where
( ) ( ) RFFwR
Iwww )1()1(1
)1(
1
)1()2(
2222
1iiii
nii
sss +−⎥⎦⎤
⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−−+= −
−
and
( ) RFFwR
Iww ni
ni
ni
nii s
ss +−⎥⎦
⎤⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−−= +
−
1
1
)1(
2.
20
3.3 Approximate Riemann Solvers Roe [12] derived an approach which approximates systems of conservation laws by
using a piecewise constant approximation
( )⎪⎪⎩
⎪⎪⎨
⎧
Δ+<<Δ−
Δ+<<Δ−=
22 if
22 if
,x
xxx
x
xxx
xx
tx
RRR
LLL
n
w
ww ,
where wL and wR represent the piecewise constant states at tn, and determining the
exact solution of a linearised Riemann problem which is related to (2.1)
( ) 0,~ =
∂∂+
∂∂
xtRL
wwwA
w, (3.6)
where ( )wF
wwA∂∂≈RL ,
~ is the linearised Jacobian matrix and ˜ is called the Roe
average. The eigenvalues and eigenvectors of A~
are λ~ and e~ respectively and are
determined from the decomposition
wAeF Δ=∑ λα=Δ=
~~~~1
k
M
kkk
where www LR −=Δ , M is the number of waves present in the system and α~
represents the wave strengths, i.e. wkk Δ=α~ . Once the eigenvalues, eigenvectors and
wave strengths associated with the linearised Riemann problem have been obtained,
Roe’s scheme [12] can be used
( )FFww *2/1
*2/1
1−+
+ −−= iini
ni s , (3.7)
where
( ) ( )( ) ⎟⎠⎞⎜⎝
⎛ ∑ −φ−λα−+== +
++
M
kkkkkk
i
ni
nii v
1 2/11
*2/1
~11~~2
1
2
1eFFF ,
λΔΔ= ~
kkx
tv ,
( )( )αα=θ
±
±
~
~
2/1
2/1
k i
k Ik , ( )( )viI k i 2/1sgn ±−=
and φk can be any of the flux-limiters listed in Table 3-1.
21
Name of Flux-limiter φ(θ) Minmod φ(θ) = max(0,min(1,θ))
Roe’s Superbee φ(θ) = max(0,min(2θ,1),min(θ,2))
van Leer ( )θ+θ+θ
=θφ1
van Albada ( )θ+θ+θ=θφ2
2
1
Table 3-1: Some second order flux-limiters.
3.3.1 Roe’s Scheme with Source Term Added
As presented so far, Roe’s scheme approximates homogeneous systems of
conservation laws. We now extend (3.7) to numerically approximate the Shallow
Water Equations, which has a source term present.
One approach is to add a source term approximation to (3.7)
( ) RFFww ss niii
ni
ni +−−= −++ *
2/1*
2/11 (S-3a)
where
( ) ( )( ) ⎟⎠⎞⎜⎝
⎛ ∑ −φ−λα−+== +
++
M
kkkkkk
i
ni
nii v
1 2/11
*2/1
~11~~2
1
2
1eFFF ,
λΔΔ= ~
kkx
tv ,
( )( )αα=θ
±
±
~
~
2/1
2/1
k i
k Ik , ( )( )viI k i 2/1sgn ±−= .
and
( )⎥⎦⎤
⎢⎣
⎡−−
=−BBhg ii
ni
ni
1
0R .
We now need to obtain the eigenvalues, eigenvectors and wave strengths associated
with the linearised Riemann problem from the decomposition. For the Shallow Water
Equations, we may obtain eigenvalues, eigenvectors and wave strengths (see
Glaister[3] and Hubbard[6])
cu ~~~1 +=λ , cu ~~~
2 −=λ ,
22
⎥⎦
⎤⎢⎣
⎡+
=cu ~~
1~1e , ⎥
⎦
⎤⎢⎣
⎡−
=cu ~~
1~2e ,
( )( )huhuc
h Δ−Δ+Δ=α ~~2
1
2
1~1
and
( )( )huhuc
h Δ−Δ−Δ=α ~~2
1
2
1~2 ,
where
hh
uhuhuLL
LLRR
++=~ and
2~ hhgc LR+= .
A semi-implicit approach of (S-3a) can be obtained by approximating the source term
at tn+1 instead of at tn, i.e.
( ) RFFww ss niii
ni
ni
1*2/1
*2/1
1 +−+
+ +−−= . (3.8)
Now, by using Taylor’s theorem, we may obtain
( ) ⎥⎦⎤
⎢⎣⎡∂∂−+≈ ++
wR
wwRR ni
ni
ni
ni
11
and by substituting into (3.8) and using (3.5), we may obtain
( ) RFFwR
Iww niii
ni
ni s
ss +−⎥⎦
⎤⎢⎣⎡
⎥⎦⎤
⎢⎣⎡∂∂−−= −+
−+ *
2/1*
2/1
1
1
2. (S-3b)
where
( ) ( )( ) ⎟⎠⎞⎜⎝
⎛ ∑ −φ−λα−+== +
++
M
kkkkkk
i
ni
nii v
1 2/11
*2/1
~11~~2
1
2
1eFFF ,
λΔΔ= ~
kkx
tv ,
( )( )αα=θ
±
±
~
~
2/1
2/1
k i
k Ik , ( )( )viI k i 2/1sgn ±−=
and
( )⎥⎦⎤
⎢⎣
⎡−−
=−BBhg ii
ni
ni
1
0R .
23
3.3.2 Roe’s Scheme with Source Term Decomposed
Instead of adding a source term approximation to Roe’s scheme [12], we can use an
approach discussed by Glaister[3]. This method approximates the source term by
decomposing the source term in a similar way that we did for the flux terms, i.e.
Re~~~1
1=∑β
Δ =k
M
kkx
.
Here, β~k are the coefficients of the decomposition of the source term onto the
eigenvectors of the characteristic decomposition (see Glaister[3] and Hubbard[6]).
For the Shallow Water Equations, we may obtain
2
~~1
HcΔ=β and 2
~~2
HcΔ−=β . (3.9)
Alternatively, we can use
2
~~1
BcΔ−=β and 2
~~2
BcΔ=β
for the Shallow Water equations. Once the values of β~k
have been obtained, we can
approximate the source term by using
( )RRR +−
−+ +
Δ= 2/12/1
* 1iii
x
where
( )( )⎟⎠⎞⎜⎝
⎛ λ±∑β== +
±+
~sgn1~~2
1 2
1 2/12/1 k
kkk
ii eR ,
which is a first order approximation of the source term. Hence, Roe’s first order
scheme with the source term decomposed may be written
( ) ( )RRFFww +−
−+−+
+ ++−−= 2/12/1*
2/1*
2/11
iiiini
ni ss (S-3c)
where
( ) ⎟⎠⎞⎜⎝
⎛ ∑ λα−+== +
++
2
1 2/11
*2/1
~~~2
1
2
1k
kkki
ni
nii eFFF
and
24
( )( )⎟⎠⎞⎜⎝
⎛ λ±∑β== +
±+
~sgn1~~2
1 2
1 2/12/1 k
kkk
ii eR .
We can also obtain a second order accurate approximation of the source term and thus
obtain Roe’s second order scheme with source term decomposed
( ) ( )RRFFww +−
−+−+
+ ++−−= 2/12/1*
2/1*
2/11
iiiini
ni ss (S-3d)
where
( ) ( ) ⎟⎠⎞⎜⎝
⎛ ∑ λα−+== +
++
2
1
2
2/11
*2/1
~~~22
1k
kkki
ni
nii
seFFF
and
( )( )⎥⎦⎤
⎢⎣⎡ λ±∑ λβ=
= +
±+
~sgn1~~~2
2
1 2/12/1 k
kkkk
ii
seR .
In addition, we can also use an approach proposed by Hubbard[6] which involves
applying TVD to the source term approximation as well as the conservation law
approximation, i.e.
( ) ( )RRFFww +−
−+−+
+ ++−−= 2/12/1*
2/1*
2/11
iiiini
ni ss (S-4)
where
( ) ( )( ) ⎟⎠⎞⎜⎝
⎛ ∑ −φ−λα−+== +
++
M
kkkkkk
i
ni
nii v
1 2/11
*2/1
~11~~2
1
2
1eFFF ,
( ) ( )( )( )⎥⎦⎤
⎢⎣⎡ −φ−λ±∑β=
= +
±+ vkkk
kkk
ii 11~sgn1~~
2
1 2
1 2/12/1 eR ,
λΔΔ= ~
kkx
tv ,
( )( )αα=θ
±
±
~
~
2/1
2/1
k i
k Ik , ( )( )viI k i 2/1sgn ±−=
and φk can be any of the flux-limiter functions listed in Table 3-1.
3.3.3 The C – Property of Bermudez & Vazquez
The advantage of decomposing the source term as well as the conservation law is that
we obtain a numerical scheme that satisfies the approximate C-property. Bermudez &
Vazquez[1] state that in order for a numerical scheme to satisfy
25
i) The approximate C-property, the numerical scheme must be at least
second order accurate when applied to the quiescent flow case, i.e. u ≡
0 and h ≡ H.
ii) The exact C-property, the numerical scheme must be exact when
applied to the quiescent flow case, i.e. u ≡ 0 and h ≡ H.
Hubbard’s approach and Roe’s scheme with source term decomposed all satisfy the
exact C-property and should produce very accurate numerical results. However,
Roe’s scheme with source term added does not even satisfy the approximate C-
property and may give misleading results as the source term becomes significant.
3.4 Q-Schemes of Bermudez & Vazquez Bermudez & Vazquez[1] discussed a variety of Q-Schemes, which numerically
approximate (1.1). All of the Q-Schemes discussed were used with the following first
order equation
( ) RFFww ts niii
ni
ni Δ+−−= −++ *
2/1*
2/11 (S-5)
where
( ) ( )( )wwwwQFFF ni
ni
ni
ni
ni
nii −−+= ++++ 111
*2/1 ,
2
1
2
1,
( ) ( )wwRwwRR ni
niiiR
ni
niiiL
ni xxxx 1111 ,,,,,, ++−− += ,
( ) ( ) ( )[ ] ( )wwRwwQwwQIwwR ni
niii
ni
ni
ni
ni
ni
niiiL xxxx ,,,ˆ,,
2
1,,, 111
1111 −−−
−−−− += ,
( ) ( ) ( )[ ] ( )wwRwwQwwQIwwR ni
niii
ni
ni
ni
ni
ni
niiiR xxxx 111
1111 ,,,ˆ,,
2
1,,, +++
−+++ −= ,
( ) ( ) ( )⎥⎥
⎦
⎤
⎢⎢
⎣
⎡
⎟⎠⎞⎜
⎝⎛
Δ−
⎟⎠⎞⎜
⎝⎛ += ++++
xxHxHhhgxx ii
ni
ni
ni
niii 1111
2
0,,,ˆ wwR
26
and Q is a matrix calculated by using a certain Q-Scheme. Bermudez & Vazquez[1]
discussed a variety of Q-Schemes but generally concentrated on the Q-scheme of van
Leer and a Q-Scheme which is equivalent to Roe’s first order scheme (S-3c).
The Q-Scheme of van Leer is
( ) ⎟⎠⎞
⎜⎝⎛ += +
+2
, 11
wwAwwQni
nin
ini ,
where A denotes the Jacobian matrix of )(wF
⎥⎦
⎤⎢⎣
⎡−
=uugh 2
10)(
2wA ,
which has eigenvalues
ghu +=λ1 and ghu −=λ2
and eigenvectors
⎥⎦
⎤⎢⎣
⎡+
=ghu
11e and ⎥
⎦
⎤⎢⎣
⎡−
=ghu
12e .
The Q-Scheme that is equivalent to Roe’s first order scheme is
( ) ( ) ⎥⎦
⎤⎢⎣
⎡−
==+uuc
ni
ni ~2~~
10~,221 wAwwQ ,
which has eigenvalues
cu ~~~1 +=λ , cu ~~~
2 −=λ
and eigenvectors
⎥⎦
⎤⎢⎣
⎡+
=cu ~~
1~1e , ⎥
⎦
⎤⎢⎣
⎡−
=cu ~~
1~2e ,
where
hh
uhuhuLL
LLRR
++=~ and
2~ hhgc LR+= .
For both Q-Schemes, the modulus of matrix A can be obtained by using
27
X�XA 1−=
where Λ represents a matrix whose diagonal elements are the eigenvalues of A,
⎥⎦
⎤⎢⎣
⎡λ
λ=2
1
0
0�
and X denotes a matrix containing the right eigenvectors,
⎥⎦
⎤⎢⎣
⎡λλ
=21
11)(wX .
Hence, for both Q-Schemes we may obtain
( ) ⎥⎦
⎤⎢⎣
⎡λλ−λλλ−λλλ
λ−λλλ−λλλ−λ
=11222112
122121
12
1A ,
and for the source term approximation we may obtain
( ) ( )( )( )
( )( )⎥⎦
⎤⎢⎣
⎡λ−λ+λ−λλλ
λλ−λλλλλ−λ
=−−
−−−−
121212
1221
2/11212 2/1
11211
,,,ˆ,,,ˆ
ii
ni
niiin
iniiiL
xxxx
wwRwwR
and
( ) ( )( )( )
( )( )⎥⎦
⎤⎢⎣
⎡λ+λ−λ−λλλ
λλ−λλ−
λλλ−λ=
++
++++
121212
1221
2/11212 2/1
11211
,,,ˆ,,,ˆ
ii
ni
niiin
iniiiR
xxxx
wwRwwR
where
( ) ( ) ( ) ( )( )xHxHhhx
gxx ii
ni
ni
ni
niii −+
Δ= ++++ 11112 2
,,,ˆ wwR .
Again, notice that (S-5) with either Q-Scheme satisfies the exact C-property.
In this chapter, we have discussed a variety of numerical schemes which approximate
(1.1). In the next chapter, we will apply these schemes to the four test problems
discussed in Chapter 2 to find out which approach is the most accurate.
28
4 Numerical Results
In this chapter, we will apply the schemes discussed in Chapter 3 and listed in Table
4-1 to the four test problems discussed in Chapter 2 to find out which approach
produces the most accurate results. We will not discuss the results of the semi-
implicit approaches as they produced almost identical results to the explicit
approaches. Also, the numerical results of Bermudez & Vazquez’s Q-Schemes will
not be discussed as the two Q-Schemes produced almost identical results to Roe’s first
order scheme with source term decomposed, i.e. (S-3b).
For the first three test problems, step-sizes Δx = 0.001 and Δt = 0.0001 will be used
with a final time of t = 0.1. A comparison will be made at t = 0.1 and numerical
results will also be shown for t = 0.01m, where m = 0 to 10.
For test Problem D, step-sizes Δx = 2800 and Δt = 1 will be used with a bed length of
L = 648,000m and a final time of t = 10,800s. A comparison will be made at t =
10,800s and numerical results will also be shown for t = 1080m, where m = 0 to 10.
Name Of Approach Reference No. Order PaperLax-Friedrichs (S-1) 1 -
MacCormack (S-2b) 2Yee[16],
LeVeque & Yee[9],Roe’s Scheme with Source Term
‘added’(S-3a) 1 / 2
Roe’s Scheme with Source TermDecomposed
(S-3b) 1
Hubbard[6],
Glaister[3],
Roe[12]
Hubbard’s Approach (S-4) 2 Hubbard[6]Table 4-1: Different approaches for numerically approximating (1.1).
29
Only first order and flux-limited second order numerical results will be discussed and
the Minmod flux-limiter will be used with all flux-limited second order approaches.
4.1 First Order Comparison
4.1.1 Numerical Results for Problem A
For this test problem, no source term is present so schemes (S-3a) and (S-3b) are the
same. Now, by using schemes (S-1) and (S-3b) to approximate Problem A and
comparing with the exact solution, we may obtain the numerical results in Figure 4-1a
and Figure 4-1b. Here, we can see that Roe’s scheme is more accurate than the Lax-
Friedrichs approach since the Lax-Friedrichs approach is more dissipative than Roe’s
scheme. Also, the Lax-Friedrichs scheme suffered badly from oscillations if larger
step-sizes were used whereas Roe’s scheme remained accurate but became more
dissipative.
4.1.2 Numerical Results for Problem B
For this test problem, a source term is now present so approaches (S-3a) and (S-3b)
are no longer equivalent. By using schemes (S-1), (S-3a) and (S-3b) to approximate
Problem B, the results in Figure 4-2a to Figure 4-3b were obtained. Here we can see
that Roe’s scheme with source term decomposed has produced the most accurate
results. Roe’s scheme with source term added produced almost identical results to
Roe’s scheme with source term decomposed showing that, for Problem B, adding a
source term approximation to Roe’s scheme produces sufficiently accurate results.
The Lax-Friedrichs approach produced the least accurate results suffering badly from
dissipation and the approach also misplaced the disturbance caused by the riverbed.
30
4.1.3 Numerical Results for Problem C
For this test problem, the source term is becoming more significant, i.e. the variation
in the riverbed is becoming more pronounced, which may cause some schemes to
produce inaccurate results. By using approaches (S-1), (S-3a) and (S-3b) to
approximate Problem C, the results in Figure 4-4a to Figure 4-7b were obtained.
Here, we can see that Roe’s scheme with source term decomposed has produced the
most accurate results but the results are no longer almost identical to Roe’s scheme
with source term added. Adding the source term in this case has produced movement
for 0.3 > x > 0.55 whereas decomposing the source term has produced no movement
for 0.3 > x > 0.55. This is because Roe’s scheme with source term decomposed
satisfies the exact C-property whereas Roe’s scheme with source term added does not.
The Lax-Friedrichs approach has produced the least accurate results due to the
scheme suffering badly from dissipation and the scheme has also produced more
movement than Roe’s scheme with source term added for 0.3 > x > 0.55. Also, from
Figure 4-7b we can see that the Lax-Friedrichs approach has started producing
oscillations at the peak of the pulse even though small step-sizes have been used.
4.1.4 Numerical Results for Problem D
For this test problem, the source term is very difficult to approximate accurately
which may cause some approaches to produce very inaccurate numerical results. By
applying schemes (S-1), (S-3a) and (S-3b) to Problem D, the results in Figure 4-8a to
Figure 4-11b were obtained. Here, we can see that Roe’s scheme with source term
decomposed has produced the most accurate results since it was the only approach not
to produce movement after x = 216,000m at t = 10,800s. This is because the
numerical scheme satisfies the exact C-property whereas the other approaches do not
even satisfy the approximate C-property. Roe’s scheme with source term added was
31
the second most accurate due to the approach producing movement after x = 216,000
at t = 10,800s. The Lax-Friedrichs approach failed to accurately approximate
Problem D at all due to the scheme producing oscillations over the whole domain.
4.1.5 Overall Comparison of First Order Approaches
From the results of this sub-section, we have seen that Roe’s scheme with source term
decomposed has produced very accurate results for all test problems suffering only
slightly from dissipation. Roe’s scheme with source term added produced accurate
numerical results for the first two test problems, but as the source term became more
significant, the numerical scheme started to produce less accurate results. In Problem
D, Roe’s scheme with source term added produced movement after x = 216,000 at t =
10,800s making the scheme very inaccurate. The Lax-Friedrichs approach is accurate
for the most basic test problems but only when sufficiently small step-sizes are used
otherwise the scheme suffers from oscillations. Also, as the source term became
significant the Lax-Friedrichs approach became impractical suffering badly from
oscillations even when small step-sizes were used. Hence, Roe’s scheme with source
term decomposed produced the most accurate results for all test problems in the first
order case since it was the only approach to satisfy the exact C-property.
4.2 Flux-Limited Second Order Comparison
4.2.1 Numerical Results for Problem A
Now, by using approaches (S-2b) and (S-4) to approximate Problem A, the results in
Figure 4-12a and Figure 4-12b were obtained. Here, we can see that Roe’s scheme
and the MacCormack approach have produced almost identical results. Both
approaches have produced extremely accurate numerical results but Roe’s scheme
was the most accurate.
32
4.2.2 Numerical Results for Problem B
By using approaches (S-2), (S-3a) and (S-4) to approximate Problem B, the results in
Figure 4-13a to Figure 4-14b were obtained. Here, we can see that Roe’s scheme with
source term added, Hubbard’s approach and the MacCormack approach have all
produced almost identical results. All three approaches have produced very accurate
results.
4.2.3 Numerical Results for Problem C
By using approaches (S-2), (S-3a) and (S-4) to approximate Problem C, the results in
Figure 4-15a to Figure 4-18b were obtained. Here, we can see that Hubbard’s
approach has produced the most accurate results. The other two approaches have
produced movement in the region 0.3 > x > 0.55 with the MacCormack approach
producing the most amount of movement of the three approaches. If we use a larger
step-size of Δx = 0.01 and Δt = 0.001 with approaches (S-2), (S-3a) and (S-4), we may
obtain the results in Figure 4-19a and Figure 4-19b. Here we can see that Hubbard’s
approach has again produced the most accurate numerical results and has not
produced movement in the region 0.3 > x > 0.55. However, the MacCormack
approach and Roe’s scheme with source term added have both produced very
inaccurate numerical results. Roe’s scheme with source term added has produced
inaccurate numerical results due to the approach producing a considerable amount of
movement in the region 0.3 > x > 0.55 and the approach has slightly smeared the
pulse at x = 0.5. The MacCormack approach has produced less movement in the
region 0.3 > x > 0.55 than Roe’s scheme with source term added but the approach has
smeared the pulse at x = 0.5 considerably more than Roe’s scheme with source term
added. Hence, Hubbard’s approach is still very accurate when the step-size is
increased but the other two approaches are becoming impractical.
33
4.2.4 Numerical Results for Problem D
Now, by using approaches (S-2), (S-3a) and (S-4) to approximate Problem D, the
results in Figure 4-20a to Figure 4-23b were obtained. Here, we can see that
Hubbard’s approach was the only approach that did not produce movement after x =
216,000m at t = 10,800s. Roe’s scheme with source term added and LeVeque &
Yee’s MacCormack approach produced very similar results but both produced
movement after x = 216,000m at t = 10,800s. Hence, Hubbard’s produced the most
accurate results.
4.2.5 Overall Comparison of the Flux- Limited Second
Order Approaches
From the results of this sub-section, we have seen that Hubbard’s approach has
produced the most accurate results for all test problems producing very accurate
results even for Problem D. LeVeque & Yee’s MacCormack approach, Roe’s scheme
with source term added and Hubbard’s approach all produced almost identical results
for the first two test problems. However, as the source term became more significant,
LeVeque & Yee’s MacCormack approach and Roe’s scheme with source term added
produced less accurate results but still produced very similar results due to both
approaches adding a source term approximation on. When the source term became
significant and a larger step-size was used, the MacCormack approach and Roe’s
scheme with source term added both became impractical but Hubbard’s approach still
produced very accurate numerical results.
34
First Order Numerical Results:
Comparison of the Different First Order Results with the Exact Solution using hx = 0.001, ht = 0.0001 and at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
h(x,
0.1)
Exact Lax-Friedrichs Roe's Scheme
Figure 4-1a: Comparison of the first order results for Problem A
Com parison of the Different First Order Results w ith the Exact Solution using hx = 0.001, ht = 0.0001 and at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
0.1)
Exact Lax-Friedrichs Roe's Schem e
Figure 4-1b: Comparison of the results for Problem A
35
Roe's Scheme with Source Term Decomposed and hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-2a: Results of Roe’s scheme with source term decomposed for Problem B
Roe's Scheme with Source Term Decomposed and hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-2b: Results of Roe’s scheme with source term decomposed for Problem B
36
Comparison of the Different First Order Approaches at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed Lax-FriedrichsRoe's Scheme with Source Term Added Roe's Scheme with Source Term Decomposed
Figure 4-3a: Comparison of the different first order approaches for Problem B
Comparison of the Different First Order Approaches at t = 0.1.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
0.1)
Lax-Friedrichs Roe's Scheme with Source Term Added Roe's Scheme with Source Term Decomposed
Figure 4-3b: Comparison of the different first order approaches for Problem B
37
Lax-Friedrichs Approach with hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-4a: Results of Lax-Friedrichs approach for Problem C
Lax-Friedrichs Approach with hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-4b: Results of Lax-Friedrichs approach for Problem C
38
Roe's Scheme w ith Source Term Decomposed and hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-5a: Results of Roe’s scheme with source term decomposed for Problem C
Roe's Scheme w ith Source Term Decomposed and hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-5b: Results of Roe’s scheme with source term decomposed for Problem C
39
Roe's Scheme w ith Source Term Added and hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-6a: Results of Roe’s scheme with source term added for Problem C
Roe's Scheme with Source Term Added and hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-6b: Results of Roe’s scheme with source term added for Problem C
40
Comparison of the Different First Order Approaches at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed Lax-FriedrichsRoe's Scheme with Source Term Added Roe's Scheme with Source Term Decomposed
0.98
1
1.02
1.04
1.06
1.08
1.1
1.12
0.3 0.35 0.4 0.45 0.5 0.55 0.6
Figure 4-7a: Comparison of the different first order approaches for Problem C
Comparison of the Different First Order Approaches at t = 0.1.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
0.1)
Lax-Friedrichs Roe's Scheme with Source Term Added Roe's Scheme with Source Term Decomposed
Figure 4-7b: Comparison of the different first order approaches for Problem C
41
Lax-Friedrichs Approach w ith hx = 1000, ht = 1 and t = 0 to 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Riverbed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-8a: Results of Lax-Friedrichs approach for Problem D
Lax-Friedrichs Approach with hx = 1000, ht = 1 and t = 0 to 10800.
-4
-3
-2
-1
0
1
2
3
4
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-8b: Results of Lax-Friedrichs approach for Problem D
42
Roe's Scheme with Source Term Decomposed and hx = 1000, ht = 1 and t = 0 to 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Riverbed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-9a: Results of Roe’s scheme with source term decomposed for Problem D
Roe's Scheme w ith Source Term Decomposed and hx = 1000, ht = 1 and t = 0 to 10800.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-9b:Results of Roe’s scheme with source term decomposed for Problem D
43
Roe's Scheme with Source Term Added and hx = 1000, ht = 1 and t = 0 to 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Seabed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-10a: Results of Roe’s scheme with source term added for Problem D
Roe's Scheme w ith the Source Term Added and hx = 1000, ht = 1 and t = 0 to 10800.
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-10b: Results of Roe’s scheme with source term added for Problem D
44
Comparison of the Different First Order Approaches at t = 10800.
0
10
20
30
40
50
60
70
0 216000 432000 648000
x
Riverbed River InitiallyLax-Friedrichs Roe's Scheme with Source Term AddedRoe's Scheme with Source Term Decomposed
Figure 4-11a: Comparison of the different first order approaches for Problem D
Comparison of the Different First Order Approaches at t = 10800.
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 216000 432000 648000
x
u(x,
1080
0)
Lax-Friedrichs Source Term Added Source Term Decomposed
-4
-3
-2
-1
0
1
2
3
4
0 216000 432000 648000x
Figure 4-11b: Comparison of the different first order approaches for Problem D
45
Flux-Limited Second Order Numerical Results:
Comparison of the Different Flux-Limited Second Order Approaches with the Exact Solution at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
h(x,
0.1)
Exact Solution MacCormack Approach Roe's Scheme
Figure 4-12a: Comparison of the flux-limited second order results for Problem A
Comparison of the Different Flux-Limited Second Order Approaches with the Exact Solution and at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
0.1)
Exact Solution MacCormack Approach Roe's Scheme
Figure 4-12b: Comparison of the flux-limited second order results for Problem A
46
Hubbard's Approach with hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-13a: Results of Hubbard’s approach for Problem B
Hubbard's Approach with hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-13b: Results of Hubbard’s approach for Problem B
47
Comparison of the Different Flux-Lim ited Second Order Approaches at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed MacCormack Approach Roe's Scheme with Source Term Added Hubbard's Approach
Figure 4-14a: Comparison of the flux-limited second order results for Problem B
Comparison of the Different Flux-Limited Second Order Appraoches at t = 0.1.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
0.1)
MacCormack Approach Roe's Scheme with Source Term Added Hubbard's Approach
Figure 4-14b: Comparison of the flux-limited second order results for Problem B
48
Hubbard's Approach with hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-15a: Results of Hubbard’s approach for Problem C
Hubbard's Approach w ith hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-15b: Results of Hubbard’s approach for Problem C
49
Roe's Scheme w ith Source Term Added and hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-16a: Results of Roe’s scheme with source term added for Problem C
Roe's Scheme w ith Source Term Added and hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-16b: Results of Roe’s scheme with source term added for Problem C
50
MacCormack Approach with hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-17a: Results of LeVeque & Yee’s MacCormack approach for Problem C
MacCormack Approach with hx = 0.001, ht = 0.0001 and t = 0 to 0.1.
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
t)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Figure 4-17b: Results of LeVeque & Yee’s MacCormack approach for Problem C
51
Comparison of the Different Flux-Limited Second Order Approaches at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Seabed MacCormack Roe's Scheme with Source Term Added Hubbard's Approach
0.98
1
1.02
1.04
1.06
1.08
1.1
1.12
0.35 0.4 0.45 0.5 0.55 0.6
Figure 4-18a: Comparison of the flux-limited second order results for Problem C
Comparison of the Different Flux-Limited Second Order Approaches at t = 0.1.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
0.1)
MacCormack Approach Roe's Scheme with Source Term Added Hubbard's Approach
Figure 4-18b: Comparison of the flux-limited second order results for Problem C
52
Comparison of the Different Flux-Limited Second Order Approaches using a Larger Step-Size of hx = 0.01, ht = 0.001 and at t = 0.1.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Riverbed MacCormack Roe's scheme with source term added Roe's scheme with source term decomposed
Figure 4-19a: Comparison of the flux-limited second order results for Problem C
Comparison of the Different Flux-Limited Second Order Approaches using a larger step size of hx = 0.01, ht = 0.001 and at t = 0.1.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
u(x,
0.1)
MacCormack Roe's scheme with source term added Hubbard's Approach
Figure 4-19b: Comparison of the flux-limited second order results for Problem C
53
Hubbard's Approach with hx = 1000, ht = 1 and t = 0 to 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Riverbed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-20a: Results of Hubbard’s approach for Problem D
Hubbard's Approach with hx = 1000, ht = 1 and t = 0 to 10800.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-20b: Results of Hubbard’s approach for Problem D
54
MacCormack Approach w ith hx = 1000, ht = 1 and t = 0 to 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Riverbed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-21a: Results of LeVeque & Yee’s MacCormack approach for Problem D
MacCormack Approach w ith hx = 1000, ht = 1 and t = 0 to 10800.
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-21b: Results of LeVeque & Yee’s MacCormack approach for Problem D
55
Roe's Scheme w ith Source Term Added and hx = 1000, ht = 1 and t = 0 to 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Seabed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-22a: Results of Roe’s scheme with source term added for Problem D
Roe's Scheme with Source Term Added and hx = 1000, ht = 1 and t = 0 to 10800.
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 4-22b: Results of Roe’s scheme with source term added for Problem D
56
Comparison of the Different Flux-Limited Second Order Approaches at t = 10800.
0
10
20
30
40
50
60
70
0 216000 432000 648000
x
Riverbed River InitiallyMacCormack Approach Roe's Scheme with Source Term AddedHubbard's Approach
Figure 4-23a: Comparison of the flux-limited second order results for Problem D
Comparison of the Different Flux-Lim ited Second Order Approaches at t = 10800.
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0 216000 432000 648000
x
u(x,
1080
0)
MacCormack Approach Roe's Scheme with Source Term Added Hubbard's Approach
-0.07
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
216000 316000 416000 516000 616000
Figure 4-23b: Comparison of the flux-limited second order results for Problem D
57
5 Conclusion
In Chapter 4, it was shown that Roe’s scheme with source term decomposed and
Hubbard’s approach produced the most accurate numerical results for all four test
problems. All the other approaches were accurate for the first two test problems, but
as the source term became more significant, the other approaches became inaccurate.
In Problem D we saw that Roe’s scheme with source term decomposed and Hubbard’s
approach was the only approach that did not produce movement after x = 216,000m at
t = 10,800s. This is because the other approaches did not satisfy the C-property of
Bermudez & Vazquez[1].
Overall, Hubbard’s approach, which is Roe’s scheme with source term decomposed
with a flux-limiter applied to both the conservation law and the source term, produced
the most accurate numerical results. However, if we use the Superbee flux-limiter
instead of the Minmod flux-limiter with Hubbard’s approach then the approach
produces inaccurate numerical results. If we use (S-4) with Roe’s Superbee flux-
limiter, the Minmod flux-limiter and van Leer’s flux-limiter and use Δx = 1000, Δt = 1
and t = 0 to 10,800s, we may obtain Figure 5-1a to Figure 5-4b. Figure 5-1a to Figure
5-2b show that by using either Roe’s Superbee flux-limiter or van Leer’s flux-limiter
with Hubbard’s approach, oscillations occur in the numerical results. Figure 5-3a and
Figure 5-3b show that using the Minmod flux-limiter with Hubbard’s approach
produces no oscillations in the numerical results. Figure 5-4a and Figure 5-4b
compare the results obtained by using the three different flux-limiters with Hubbard’s
approach at t = 10,800s. Here, we can see that Roe’s Superbee flux-limiter has
58
produced the most oscillations, followed by van Leer’s flux-limiter and the Minmod
flux-limiter has produced no oscillations. This suggests that applying a flux-limiter to
the source term approximation when the source term is significant can create
oscillations in the numerical results. However, if we do not apply a flux-limiter to the
source term approximation but we do to the conservation law then we will no longer
be able to obtain a numerical scheme which satisfies the C-property of Bermudez &
Vazquez[1]. The only solution at present is to use the Minmod flux-limiter with
Hubbard’s approach or to use Roe’s first order scheme with source term decomposed.
Throughout this report, we have seen that adding a source term approximation can
produce accurate numerical results but as the source term becomes significant, adding
a source term approximation can give very inaccurate numerical results. We have
also shown that by decomposing the source term as well as the conservation law, we
may obtain very accurate numerical results for the first order case. However, when
applying flux-limiters to the source term as well as the conservation law to ensure the
numerical scheme satisfies the exact C-property, oscillations can occur in the
numerical solution depending on which flux-limiter is used.
59
Superbee Flux-Limiter with hx = 1000, ht = 1 and t = 0 to 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Seabed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 5-1a: Superbee Flux-Limiter results for Problem D
Superbe Flux-Limiter with hx = 1000, ht = 1 and t = 0 to 10800.
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 5-1b: Superbee Flux-Limiter results for Problem D
60
van Leer Flux-Limiter with hx = 1000, ht = 1 and t = 0 to 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Seabed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 5-2a: van Leer Flux-Limiter results for Problem D
van Leer Flux-Limiter with hx = 1000, ht = 1 and t = 0 to 10800.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 5-2b: van Leer Flux-Limiter results for Problem D
61
Minmod Flux-Limiter with hx = 1000, ht = 1 and t = 0 t o 10800.
0
10
20
30
40
50
60
70
0 100000 200000 300000 400000 500000 600000
x
Seabed 0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 5-3a: Minmod Flux-Limiter results for Problem D
Minmod with hx = 1000, ht = 1 and t = 0 to 10800.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 100000 200000 300000 400000 500000 600000
x
u(x,
t)
0 1080 2160 3240 4320 5400 6480 7560 8640 9720 10800
Figure 5-3b: Minmod Flux-Limiter results for Problem D
62
Comparison of the Three Flux-Limiters at t = 10800.
0
10
20
30
40
50
60
70
0 216000 432000 648000
x
Riverbed River Initially Superbee van Leer Minmod
Figure 5-4a: Comparison of the Flux-Limiter results for Problem D
Comparison of the Three Flux-Limiters at t = 10800.
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
0 216000 432000 648000
x
u(x,
1080
0)
Superbee van Leer Minmod
Figure 5-4b: Comparison of the Flux-Limiter results for Problem D
63
Acknowledgements
I would like to thank my supervisor Dr. P.K. Sweby and Dr. M.E. Hubbard for their
guidance and support. In addition, I would like to thank the EPSRC and HR
Wallingford for their funding.
64
References1. A. Bermudez and M^Elena Vazquez, Upwind Methods for Hyperbolic
Conservation Laws with Source Terms, Computers and Fluids Vol. 23, No. 8,1049 – 1071(1994).
2. P. Garcia-Navarro, Some Considerations and Improvements on the Performanceof Roe’s scheme for 1D Irregular Geometries, Internal Report 23 (1997).
3. P. Glaister, Difference Schemes for the Shallow Water Equations, NumericalAnalysis Report 9/87, University Of Reading (1987).
4. P. Glaister, Approximate Riemann Solutions of the Shallow Water Equations, J.Hydr. Research Vol. 26, No.3 (1988).
5. A. Harten, High Resolution Schemes for Conservation Laws, J. Comput. Phys. 49(1983).
6. M.E. Hubbard, Private Communication (University of Reading).
7. J. Hudson, Numerical Techniques for Conservation Laws with Source Terms, MScDissertation, University of Reading (1998).
8. D. Kroner, Numerical Schemes for Conservation Laws, Wiley-Teubner series(1997).
9. R.J. LeVeque and H.C. Yee, A Study of Numerical Methods for HyperbolicConservation Laws with Stiff Source Terms, J. Comput. Phys. 86, 187 – 210(1990).
10. R.J. LeVeque, Numerical Methods for Conservation Laws, Birkhauser-Verlag(1990).
11. R.J. LeVeque, Balancing Source Terms and Flux Gradients in High-ResolutionGodunov Methods: The Quasi-Steady Wave-Propagation Algorithm, J. Comput.Phys. (1998).
12. P.L. Roe, Approximate Riemann Solvers, Parameter Vectors and DifferenceSchemes, J. Comput. Phys. 43, 357 – 372 (1981).
13. J.J Stoker, Water Waves, Wiley Interscience (1957).
14. P.K. Sweby, High Resolution Schemes Using Flux Limiters for HyperbolicConservation Laws, SIAM J. Num. Anal. 21, 995 (1984).
15. P.K. Sweby, Source Terms and Conservation Laws: A Preliminary Discussion,Numerical Analysis Report 6/89, University Of Reading (1989)
16. H.C. Yee, Upwind and Symmetric Shock-Capturing Schemes, NASA AmesResearch Center Technical Memoranda 89464 (1987).